Professional Documents
Culture Documents
Leader Election CRN
Leader Election CRN
Wireless Networks
1 Introduction
Currently, the wireless channels are allocated using the fixed spectrum alloca-
tion policy in which the spectrum is assigned by a central authority to various
services. However some frequencies are much more in demand than other owing
to their lower energy requirements, low error rate and higher transmission range.
As a result, a large portion of the spectrum is sporadically used whereas in some
portions the usage is a lot more concentrated [1]. In order to allocate channels
efficiently so as to avoid spectrum holes, the use of cognitive radios has been
recommended [1,2].
Cognitive radios (CRs) are able to change their transmission and reception
parameters to communicate efficiently avoiding interference with other users by
sensing the spectrum. This adjustment can be done dynamically and can in-
clude changes in communication frequency, encoding, link layer parameters etc.
Even though cognitive radios are equipped with only a single antenna, still this
dynamic adjustment allows them to operate on multiple channels.
While availability of multiple channels allow better network throughput po-
tentially, but it also makes the setting up of the network much more difficult.
This is because different nodes may have different capabilities. For example, a
node R1 may be able to communicate on channel C5 , but a neighboring node R2
may not be able to communicate on that channel because of different hardware
capability or spatial variance of the channel availability set [1]. This makes it a
lot more difficult to solve traditional problems for multi-channel radio networks.
Leader election is one such fundamental problem. It involves selecting a distin-
guished node in the network. In wireless networks, leader election has been used
to solve many other problems like routing [3], key distribution [4], coordination
in sensor networks [5], neighbor discovery [6,7] and so on.
For multi-channel networks, a leader can be elected in two different ways:
(i) one leader for all the channels, or (ii) one leader on each channel with possibly
different leaders for different channels. Having one leader for all the channels
is not feasible if no node in the network has the capability to communicate
on all the channels. Therefore, to ensure that every node in the network can
communicate with a leader of a channel in a single step, we have to ensure that
there is a leader for every channel on which one or more nodes in the network
can operate.
In this paper, we propose several leader election algorithms (one determin-
istic and two randomized) for single-hop multi-channel networks (like cognitive
radio networks). The deterministic algorithm guarantees that a leader is elected
for all channels using at most 2N + M log M time slots, where M is the size
of the universal channel set available to the nodes and N is the size of the
label space used to assign unique labels to the nodes. Note that the time com-
plexity of our deterministic algorithm is much lower than that of the naı̈ve
algorithm, which uses M N time slots. The randomized algorithms guarantee
that a leader is elected for all the channels with probability at least 1 − f1 within
O(M log2 (nmax )+M log2 (M f )) and O(M log(nmax ) log(M f )) time slots, respec-
tively, where nmax is the maximum number of nodes operating on any channel in
the network. Both randomized algorithms work without assuming the knowledge
of nmax or N . As it can be observed, the time complexity of both randomized al-
gorithms is asymptotically much lower than that of the deterministic algorithm
when N M log2 (nmax ) + M log2 (M f ).
The rest of the paper is organized as follows. In section 2, we discuss the
previous work conducted on leader election in single channel networks. We
describe our system model in section 3. In section 4 we describe a determin-
istic algorithm for leader election. In section 5, we describe two randomized
algorithms for leader election. Finally we conclude the paper in section 6. Due
to space constraints, some of the proofs have been omitted and can be found
elsewhere [8].
312 T. Bansal, N. Mittal, and S. Venkatesan
2 Related Work
Depending on radio transceiver’s capability to detect collision, two radio models
have been used in literature [9] for developing wireless algorithms viz. “Strong
Radio model” and “Weak Radio Model”. In the weak radio model, nodes cannot
distinguish between “no transmission” and “multiple transmissions”. This is also
known as Radio Network with no Collision Detection (no-CD RN). Moreover,
the transmitting nodes are not capable of monitoring the state of the channel
simultaneously while transmitting. In strong radio model, nodes have collision
detection capability which allows them to differentiate between “no transmis-
sion” and “multiple transmissions”. Strong radio model also allows transmitting
nodes to detect collision by simultaneously allowing them to monitor the channel
state. As pointed out by the authors [9], only the weak radio model is consistent
with the IEEE 802.11 standards and current WLAN standards.
For deterministic leader election algorithms for single channel weak radio net-
works, Jurdzinski et al. [10] have proved that the lower bound on time require-
ments for such algorithms is N − 1 time slots. Nakano et al. [11] have given one
such time optimal and energy efficient deterministic algorithm which terminates
in optimal number of time slots.
A trivial way to design deterministic algorithm for leader election for multiple
channels could be to run M instances of Nakano’s [11] algorithm on each channel.
However, in that case the time complexity of the protocol would become O(N M ).
In this paper, we present a deterministic algorithm for leader election which
makes use of FDMA to achieve a much lower time complexity of O(N +M log M ).
Different randomized algorithms for leader election have been proposed in the
literature for different radio models. The leader election algorithms proposed in
[10], [12], [13], [14] are based on strong radio model. They either assume that the
nodes have collision detection capability or that the nodes can monitor the state
of the channel at the same time while transmitting. In this paper, we consider
only those algorithms that assume weak radio model.
Metcalfe et al. [15] have proposed algorithms for leader election which assumes
the knowledge of the number of nodes present in the network. However for multi-
channel networks, it is possible that the number of nodes present on different
channels differ vastly. This may be due to different hardware capabilities of the
devices or because of the spatial variance of the channel availability set [1]. For
example, a higher percentage of nodes may be able to communicate on channels
which have low energy requirements, low error rate etc. as compared to channels
which do not exhibit these properties. This makes it very difficult to predict the
number of nodes on each channel as required by their algorithm.
Hayashi et al. [16] have given a randomized algorithm for leader election for the
weak radio model with unknown number of nodes present on the channel, which
terminates in O(log2 n) broadcast rounds with probability exceeding 1 − 1/n
where n is the actual number of nodes present on the channel. Their algorithm
proceeds in multiple rounds and each round is further divided into time slots. The
number of time slots in a round increase with the round number. The ith round
of execution is divided into i time slots. At the beginning of each round, nodes
Leader Election Algorithms for Multi-channel Wireless Networks 313
set the probability of transmission as 1/2 and with each time slot in that round,
the probability of transmission is reduced by half. The algorithm terminates as
soon as a time slot occurs in which exactly a single node transmits.
For the networks where the number of radio nodes present is not known,
Nakano et al. [11] have also proposed three different algorithms for leader elec-
tion. The first algorithm proposed by them terminates with probability exceeding
1−1/f in O((log2 n)+(log2 f )) time slots. One of our randomized leader election
algorithm is derived from this algorithm.
3 System Model
We consider leader election problem for one hop radio network with weak radio
model. We term the set of all the channels as universal channel set, Auniv . The
nodes can communicate on only a subset of the channels of the global channel
set. So two nodes Ri and Rj would be termed as neighbors only if there exists
some channel Ck such that both Ri and Rj can communicate on Ck . We assume
that the nodes have been assigned unique labels from the set: {1, 2, 3, ..., N }.
In this paper we give a deterministic and a randomized algorithm for leader
election in multi channel networks. Unless otherwise stated, we assume that all
nodes are aware of the global channel set and the size of the label space. We use
the following notation to describe our algorithms:
- M = Number of channels in the universal channel set
- Auniv = Universal channel set
- N = Size of the label space assigned to the nodes
- Ci = Channel i from the universal channel set Auniv
- Ri = Radio node with label i
- Ai = Set of channels available to node Ri , if Ri is present in the network
- nmax = Maximum number of nodes present on any channel
Observe that N acts as an upper bound on nmax . However, in practice, nmax
may be much smaller than N . Also, observe that there may be one or more
channels on which no node can operate. Clearly, a leader cannot be elected for
such channels. Therefore we call a channel relevant if there are one or more
nodes in the network that can operate on that channel.
N
Round 1 ··· Round j ··· Round 2i M
M time slots
Figure 1 gives the high level overview of the working of the algorithm. The
execution of the algorithm is divided into log N phases and each phase is further
divided into k rounds where k depends on the label space size at the beginning
of the round. Further, each round takes M time slots to complete.
Working within a phase: Let us assume that at the beginning of the phase, the
size of the label space of the competing nodes is X. So we divide this label space
into groups of size 2M . There are 2M
X
such groups. We then further divide the
execution of the phase into 2M rounds.
X
In each round, the algorithm handles one block of label space of size 2M . In
each round, we reduce the size of label space of competing nodes from 2M to
M . So if at the start of the round, all the label space was occupied by nodes
such that there were up to 2M nodes competing to become leader, we ensure
that when the execution of the round terminates, out of the 2M nodes at most
M nodes are still qualified for being elected as leaders. Thereafter, only these
M nodes, get the chance to participate in the leader election process in the next
phase. The remaining M nodes are no longer eligible for leader election on that
channel.
However, if at the beginning of the phase, number of competing nodes is less
than 2M , then the phase would complete in a single round. Note that in this
case as well, the algorithm would be able to reduce the label space size by a
factor of 2.
Table 1. Shows the execution of the algorithm within a round when label space of
participating nodes is {1, 2, ..., 2M }. The addition is done modulo M . In case there
are some labels with no nodes, then the time slots corresponding to those labels remain
blank.
Channel
C C2 ... Cj ... CM −1 CM
Time Slot 1
1 1, 2 3, 4 2j − 1, 2j 2M − 2M − 1, 2M
3, 2M − 2
2 2M − 1, 2M 1, 2 2j − 3, 2j − 2 2M − 5, 2M − 3,
2M − 4 2M − 2
i 2M − 2i + 3, 2M − 2i + 5, 2M −2i+2j +1, 2M − 2i − 1, 2M − 2i + 1,
2M − 2i + 4 2M − 2i + 6 2M − 2i+ 2j + 2 2M − 2i 2M − 2i + 2
M −1 5, 6 7, 8 2j + 3, 2j + 4 1, 2 3, 4
M 3, 4 5, 6 2j + 1, 2j + 2 2M − 1, 2M 1, 2
Lemma 1. On any channel, no two nodes which are competing to become leader
on that channel can have the same label.
X
N
×M (1)
i=1
2 ×M
i
where X is the number of phases in the execution. For values of N of the form
M × 2t (where t ∈ N), X = t + log M . The first t phases reduce the label
space size from N to M and the remaining log M phases reduce the label
space size from M to 1. It can be verified that this summation evaluates to
N − M + M log M . For general values of N with N > M which are not of this
form, we pick the smallest integer N of the form M × 2t (where t ∈ N) that is at
least N . This new integer N is then considered as the new N . The time required
for completion would be N − M + M log M time slots. Since N ≤ 2N , the
execution of the deterministic protocol takes at most 2N − M + M log M time
slots.
Finally, for values of N and M such that N ≤ M , it can be shown that
summation simplifies to M log N , which is at most M log M .
Leader Election Algorithms for Multi-channel Wireless Networks 317
Table 2. Description of the execution of the protocol over the multiple phases. Column
2 shows the range of original labels possible for the nodes having the corresponding
labels in Column 1 during the particular phase. (a) Phase 1 execution: Six labels (or 3
rows of the table) are processed in 1st round of execution. Only 4 labels are processed in
the 2nd round. (b) Phase 2 execution, (c) Phase 3 execution, and (d) Phase 4 execution.
15 time slots. Note that if we had elected leaders the trivial way, the total time
slots required would have been N M = 10 ∗ 3 = 30 time slots.
Theorem 2. Let nmax denote the maximum number of nodes that can be present
on any channel in the network. Then Rand-Elect elects leader for all relevant
channels within O(M log2 (nmax ) + M log2 (M f )) time slots with probability at
least 1 − f1 .
Algorithm 2. Alg-Known-Size(S, T)
t←1
repeat
transmit with probability S1
t ← t+1
until exactly one node transmits or t > T
Theorem 3. Let nmax denote the maximum number of nodes present on any
channel in the network. Then Fast-Rand-Elect elects a leader on all relevant
channels with probability exceeding 1 − f1 within O(M log(nmax ) log(M f )) time
slots.
Observe that Fast-Rand-Elect has better asymptotic time complexity than Rand-
Elect if log(nmax ) is either o(log(M f )) or ω(log(M f )). In other words, log(nmax )
∈ Θ(log(M f )). In case log(nmax ) = Θ(log(M f )), the two algorithms have the
same asymptotic time complexity. On the other hand, Fast-Rand-Elect requires
a user to fix the success probability (given by 1 − 1/f ) a priori.
6 Conclusion
In this paper, we have presented three leader election algorithms for multi chan-
nel radio networks with no collision detection. The deterministic algorithm guar-
antees leader election for all relevant channels (available to one or more nodes in
the network) and takes at most 2N +M log M time slots, which is a considerable
improvement over the trivial algorithm which takes N M time slots for comple-
tion. The two randomized algorithms guarantee leader election for all relevant
channels with probability at least 1 − f1 within O(M log2 (nmax ) + M log2 (M f ))
and O(M log(nmax ) log(M f )) time slots, respectively.
Note that, when nmax = Θ(M ), the deterministic algorithm has time-
complexity of O(M log M ), which is asymptotically better than that of the ran-
domized algorithms. Moreover, with a randomized algorithm, there is always a
non-zero probability that a leader may not be elected for one or more of the rel-
evant channels. However, the deterministic algorithm guarantees that a leader
is elected for all the relevant channels with 100% confidence.
Leader Election Algorithms for Multi-channel Wireless Networks 321
References
1. Akyildiz, I.F., Lee, W.Y., Vuran, M.C., Mohanty, S.: Next generation/dynamic
spectrum access/cognitive radio wireless networks: A survey. Computer Net-
works 50(13), 2127–2159 (2006)
2. Mitola, J.: Cognitive radio: An Integrated Agent Architecture for Software Defined
Radio. PhD thesis, Royal Institute of Technology (2000)
3. Perkins, C.E., Belding-Royer, E.M.: Ad-hoc on-demand distance vector routing. In:
Proceedings of the 2nd Workshop on Mobile Computing Systems and Applications
(WMCSA), pp. 90–100. IEEE Computer Society, Los Alamitos (1999)
4. DeCleene, B., Dondeti, L., Griffin, S., Hardjono, T., Kiwior, D., Kurose, J., Towsley,
D., Vasudevan, S., Zhang, C.: Secure group communication for wireless networks.
In: Proceedings of the IEEE Military Communications Conference (MILCOM), pp.
113–117 (2001)
5. Heinzelman, W.R., Chandrakasan, A., Balakrishnan, H.: Energy-efficient commu-
nication protocol for wireless microsensor networks. In: Proceedings of the Annual
Hawaii International Conference on System Sciences (HICSS) (2000)
6. Mittal, N., Krishnamurthy, S., Chandrasekaran, R., Venkatesan, S.: A Fast Deter-
ministic Algorithm for Neighbor Discovery in Multi-Channel Cognitive Radio Net-
works. Technical Report UTDCS-14-07, The University of Texas at Dallas (2007)
7. Krishnamurthy, S., Thoppian, M.R., Kuppa, S., Chandrasekaran, R., Mittal, N.,
Venkatesan, S., Prakash, R.: Time-efficient distributed layer-2 auto-configuration
for cognitive radio networks. Computer Networks 52(4), 831–849 (2008)
8. Bansal, T., Mittal, N., Venkatesan, S.: Leader Election Algorithms for Multi-
Channel Wireless Networks. Technical Report UTDCS-19-08, The University of
Texas at Dallas (2008)
9. Jurdzinski, T., Kutylowski, M., Zatopianski, J.: Weak communication in single-
hop radio networks: Adjusting algorithms to industrial standards. Concurrency
and Computation: Practice and Experience 15(11-12), 1117–1131 (2003)
10. Jurdzinski, T., Kutylowski, M., Zatopianski, J.: Efficient algorithms for leader elec-
tion in radio networks. In: Proceedings of the 21st ACM Symposium on Principles
of Distributed Computing (PODC), pp. 51–57 (2002)
11. Nakano, K., Olariu, S.: Randomized leader election protocols in radio networks
with no collision detection. In: Lee, D.T., Teng, S.-H. (eds.) ISAAC 2000. LNCS,
vol. 1969, pp. 362–373. Springer, Heidelberg (2000)
12. Nakano, K., Olariu, S.: Uniform leader election protocols for radio networks. IEEE
Trans. Parallel Distrib. Syst. 13(5), 516–526 (2002)
13. Bordim, J.L., Ito, Y., Nakano, K.: An energy efficient leader election protocol for
radio network with a single transceiver. IEICE Transactions 89-A(5), 1355–1361
(2006)
14. Nakano, K., Olariu, S.: A survey on leader election protocols for radio networks. In:
Proceedings of the International Symposium on Parallel Architectures, Algorithms
and Networks (ISPAN), pp. 63–68. IEEE Computer Society, Los Alamitos (2002)
15. Metcalfe, R., Boggs, D.: Ethernet: Distributed packet switching for local computer
networks. Commun. ACM 19(7), 395–404 (1976)
16. Hayashi, T., Nakano, K., Olariu, S.: Randomized initialization protocols for packet
radio networks. In: Proceedings of the 13th International Parallel and Processing
Symposium (IPPS), pp. 544–548. IEEE Computer Society, Los Alamitos (1999)